anzorq's picture
Update README.md
c74922e
metadata
language:
  - kbd
  - ru
  - multilingual
license: unknown
tags:
  - circassian
  - kabardian
datasets:
  - anzorq/kbd_lat-835k_ru-3M

t5-v1_1-small pretrained with mlm task on

• kbd (custom latin script) 835K lines: a pile of scraped text from news sites, books etc.

• ru 3M lines: wiki corpus from OPUS

tokenizer: sentencepiece unigram, 8K, shared vocabulary